[flang-commits] [flang] [flang][cuda] Unify registration under -gpu=mem:unified (PR #212871)

[Zhen Wang via flang-commits]

================
@@ -31,21 +31,21 @@ module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<!llvm.ptr, dense<
 // CHECK: gpu.module @cuda_device_mod
 
 // CHECK: llvm.func internal @__cudaFortranConstructor() {
-// CHECK-DAG: %[[MODULE:.*]] = cuf.register_module @cuda_device_mod -> !llvm.ptr
-// CHECK-DAG: %[[VAR_NAME:.*]] = fir.address_of(@_QQ{{.*}}) : !fir.ref<!fir.char<1,12>>
-// CHECK-DAG: %[[VAR_ADDR:.*]] = fir.address_of(@_QMmtestsEn) : !fir.ref<!fir.array<5xi32>>
-// CHECK-DAG: %[[MODULE2:.*]] = fir.convert %[[MODULE]] : (!llvm.ptr) -> !fir.ref<!fir.llvm_ptr<i8>>
-// CHECK-DAG: %[[VAR_ADDR2:.*]] = fir.convert %[[VAR_ADDR]] : (!fir.ref<!fir.array<5xi32>>) -> !fir.ref<i8>
-// CHECK-DAG: %[[VAR_NAME2:.*]] = fir.convert %[[VAR_NAME]] : (!fir.ref<!fir.char<1,12>>) -> !fir.ref<i8>
-// CHECK-DAG: %[[CST:.*]] = arith.constant 20 : index
-// CHECK-DAG: %[[CST2:.*]] = fir.convert %[[CST]] : (index) -> i64
-// CHECK-DAG: fir.call @_FortranACUFRegisterVariable(%[[MODULE2]], %[[VAR_ADDR2]], %[[VAR_NAME2]], %[[CST2]]) : (!fir.ref<!fir.llvm_ptr<i8>>, !fir.ref<i8>, !fir.ref<i8>, i64) -> ()
-// CHECK-DAG: %[[BOX:.*]] = fir.address_of(@_QMmtestsEndev) : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
-// CHECK-DAG: %[[BOXREF:.*]] = fir.convert %[[BOX]] : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) -> !fir.ref<i8>
-// CHECK-DAG: fir.call @_FortranACUFRegisterVariable(%[[MODULE:.*]], %[[BOXREF]], %{{.*}}, %{{.*}})
+// NOUNIFIED-DAG: %[[MODULE:.*]] = cuf.register_module @cuda_device_mod -> !llvm.ptr
+// NOUNIFIED-DAG: %[[VAR_NAME:.*]] = fir.address_of(@_QQ{{.*}}) : !fir.ref<!fir.char<1,12>>
+// NOUNIFIED-DAG: %[[VAR_ADDR:.*]] = fir.address_of(@_QMmtestsEn) : !fir.ref<!fir.array<5xi32>>
+// NOUNIFIED-DAG: %[[MODULE2:.*]] = fir.convert %[[MODULE]] : (!llvm.ptr) -> !fir.ref<!fir.llvm_ptr<i8>>
+// NOUNIFIED-DAG: %[[VAR_ADDR2:.*]] = fir.convert %[[VAR_ADDR]] : (!fir.ref<!fir.array<5xi32>>) -> !fir.ref<i8>
+// NOUNIFIED-DAG: %[[VAR_NAME2:.*]] = fir.convert %[[VAR_NAME]] : (!fir.ref<!fir.char<1,12>>) -> !fir.ref<i8>
+// NOUNIFIED-DAG: %[[CST:.*]] = arith.constant 20 : index
+// NOUNIFIED-DAG: %[[CST2:.*]] = fir.convert %[[CST]] : (index) -> i64
+// NOUNIFIED-DAG: fir.call @_FortranACUFRegisterVariable(%[[MODULE2]], %[[VAR_ADDR2]], %[[VAR_NAME2]], %[[CST2]]) : (!fir.ref<!fir.llvm_ptr<i8>>, !fir.ref<i8>, !fir.ref<i8>, i64) -> ()
+// NOUNIFIED-DAG: %[[BOX:.*]] = fir.address_of(@_QMmtestsEndev) : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
+// NOUNIFIED-DAG: %[[BOXREF:.*]] = fir.convert %[[BOX]] : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) -> !fir.ref<i8>
+// NOUNIFIED-DAG: fir.call @_FortranACUFRegisterVariable(%[[MODULE:.*]], %[[BOXREF]], %{{.*}}, %{{.*}})
 // Device and constant globals already live in device memory, so they are not
 // statically registered as device-resident under -gpu=mem:unified.
----------------
wangzpgi wrote:

Is this comment still true?

https://github.com/llvm/llvm-project/pull/212871